Improving River Routing Using a Differentiable Kinematic Wave Model and Physics-Informed Machine Learning

This project aims to improve hydrological modeling by combining Machine Learning (ML) models with physics-based models to better simulate how water flows through river networks. While traditional models are physically consistent but limited in using big data, and ML models are powerful but lack interpretability, this project integrates both approaches through a framework called differentiable modeling (DM). The work will focus on enhancing the HYDROTEL modeling platform by improving its river routing component and extending it to water temperature module as well. A pilot study on the Nicolet watershed, Quebec, Canada, will be conducted to evaluate the approach. The expected outcomes are anticipated to improve the accuracy of streamflow and water temperature simulations, supporting better water resource management and decision-making under climate change. The proposed DM framework will enhance knowledge discovery at the participating institutions by advancing hybrid modeling and integrating it with large-scale HYDROTEL simulations across the hydroclimatic atlas of southern Québec.

Faculty Supervisor:

Alain N Rousseau

Student:

Partner:

Indian Institute of Technology Bombay

Discipline:

Engineering

Sector:

Education

University:

Université du Québec : Institut national de la recherche scientifique

Program: